Alex Wissner-Gross: A new equation for intelligence

Prevodilac: Miloš Milosavljević Lektor: Mile Živković

Intelligence -- what is it?

Inteligencija - šta je to?

If we take a look back at the history

Ako pogledamo unazad kroz istoriju

of how intelligence has been viewed,

kako su videli inteligenciju,

one seminal example has been

jedan važan primer je bio

Edsger Dijkstra's famous quote that

poznati citat Edsgara Dajkstre:

"the question of whether a machine can think

"Pitanje da li mašina može da misli

is about as interesting

je isto tako zanimljivo

as the question of whether a submarine

kao i pitanje da li podmornica

can swim."

može da pliva."

Now, Edsger Dijkstra, when he wrote this,

Kad je Edsgar Dajkstra napisao ovo,

intended it as a criticism

nameravao je da to bude kritika

of the early pioneers of computer science,

pionira u kompjuterskoj nauci

like Alan Turing.

kao što je Alan Tjuring.

However, if you take a look back

Međutim, ako pogledate u prošlost

and think about what have been

i razmislite koje su bile

the most empowering innovations

najvažnije inovacije

that enabled us to build

koje su nam omogućile da pravimo

artificial machines that swim

veštačke mašine koje plivaju

and artificial machines that [fly],

i veštačke mašine koje lete,

you find that it was only through understanding

otkrićete da smo samo pomoću razumevanja

the underlying physical mechanisms

fizičkih mehanizama koji čine osnovu

of swimming and flight

plivanja i letenja

that we were able to build these machines.

mogli da napravimo ove mašine.

And so, several years ago,

I tako, pre nekoliko godina,

I undertook a program to try to understand

sproveo sam program da bih pokušao da razumem

the fundamental physical mechanisms

fizičke mehanizme

underlying intelligence.

koji čine osnovu inteligencije.

Let's take a step back.

Hajde da se vratimo korak unazad.

Let's first begin with a thought experiment.

Da počnemo prvo sa misaonim eksperimentom.

Pretend that you're an alien race

Zamislite da ste vanzemaljska rasa

that doesn't know anything about Earth biology

koja ne zna ništa o biologiji Zemlje

or Earth neuroscience or Earth intelligence,

ili o neuronauci Zemlje ili o Zemljinoj inteligenciji,

but you have amazing telescopes

ali imate neverovatne teleskope

and you're able to watch the Earth,

i možete da posmatrate Zemlju

and you have amazingly long lives,

i imate neverovatno duge živote,

so you're able to watch the Earth

tako da možete da posmatrate Zemlju

over millions, even billions of years.

milonima ili čak milijardama godina.

And you observe a really strange effect.

I primećujete veoma čudan efekat.

You observe that, over the course of the millennia,

Primećujete da je tokom milenijuma

Earth is continually bombarded with asteroids

Zemlja konstantno bombardovana asteroidima

up until a point,

sve do jednog momenta

and that at some point,

i da u jednom momentu

corresponding roughly to our year, 2000 AD,

koji odgovara, po gruboj proceni, našoj 2000. godini,

asteroids that are on

asteroidi koji su na putu

a collision course with the Earth

da se sudare sa Zemljom,

that otherwise would have collided

koji bi inače udarili,

mysteriously get deflected

misteriozno bivaju skrenuti

or they detonate before they can hit the Earth.

ili eksplodiraju pre nego što udare u Zemlju.

Now of course, as earthlings,

Naravno, kao Zemljani

we know the reason would be

znamo da je razlog to

that we're trying to save ourselves.

što pokušavamo da se spasemo.

We're trying to prevent an impact.

Pokušavamo da sprečimo udarac.

But if you're an alien race

Ali ako ste vanzemaljska rasa

who doesn't know any of this,

koja ne zna ništa od ovoga,

doesn't have any concept of Earth intelligence,

nema nikakvu ideju o inteligenciji na Zemlji,

you'd be forced to put together

bili biste prisiljeni da smislite

a physical theory that explains how,

fizičku teoriju koja objašnjava kako

up until a certain point in time,

do određenog trenutka u vremenu

asteroids that would demolish the surface of a planet

asteroidi koji bi uništili površinu planete

mysteriously stop doing that.

misteriozno ne učine to.

And so I claim that this is the same question

Tvrdim da je ovo isto pitanje

as understanding the physical nature of intelligence.

kao i razumevanje fizičke prirode inteligencije.

So in this program that I undertook several years ago,

U ovom programu koji sam sproveo pre nekoliko godina

I looked at a variety of different threads

video sam raznovrsne tragove

across science, across a variety of disciplines,

u nauci, u raznovrsnim disciplinama

that were pointing, I think,

koji su ukazivali, ja mislim,

towards a single, underlying mechanism

na jedan mehanizam koji je u osnovi

for intelligence.

inteligencije.

In cosmology, for example,

U kosmologiji, na primer,

there have been a variety of different threads of evidence

bilo je mnoštvo različitih tragova dokaza

that our universe appears to be finely tuned

da je naš svemir, izgleda, lepo naštimovan

for the development of intelligence,

za razvoj inteligencije

and, in particular, for the development

i posebno, za razvoj

of universal states

univerzalnih stanja

that maximize the diversity of possible futures.

koja maksimiziraju raznolikost mogućih budućnosti.

In game play, for example, in Go --

U igrama na primer, u igri Go -

everyone remembers in 1997

svi se sećaju kada je 1997.

when IBM's Deep Blue beat Garry Kasparov at chess --

IBM-ov "Dip blu" pobedio Garija Kasparova u šahu -

fewer people are aware

manje ljudi je svesno činjenice

that in the past 10 years or so,

da je u poslednjih 10 godina,

the game of Go,

igra Go,

arguably a much more challenging game

možda mnogo izazovnija igra od šaha

because it has a much higher branching factor,

jer ima mnogo veći faktor grananja,

has also started to succumb

takođe počela da se predaje

to computer game players

pred igračima kompjuterskih igara

for the same reason:

iz istog razloga:

the best techniques right now for computers playing Go

trenutno, najbolje tehnike za kompjutere koji igraju Go

are techniques that try to maximize future options

su tehnike koje pokušavaju da maksimiziraju buduće opcije

during game play.

tokom igre.

Finally, in robotic motion planning,

Na kraju, u planiranju robotskih pokreta

there have been a variety of recent techniques

postoji mnoštvo novih tehnika

that have tried to take advantage

koje pokušavaju da iskoriste

of abilities of robots to maximize

mogućnosti robota da maksimiziraju

future freedom of action

buduću slobodu delovanja

in order to accomplish complex tasks.

da bi obavili složene zadatke.

And so, taking all of these different threads

I tako, ako uzmemo sve ove različite tragove

and putting them together,

i sastavimo ih,

I asked, starting several years ago,

počeo sam da pitam pre nekoliko godina,

is there an underlying mechanism for intelligence

da li postoji mehanizam u osnovi inteligencije

that we can factor out

koji možemo da izdvojimo

of all of these different threads?

od svih ovih različitih tragova?

Is there a single equation for intelligence?

Da li postoji jedna jednačina za inteligenciju?

And the answer, I believe, is yes. ["F = T ∇ Sτ"]

I odgovor je, ja mislim, da. ["F = T ∇ Sτ"]

What you're seeing is probably

Ono što vidite je verovatno

the closest equivalent to an E = mc²

najbliži ekvivalent E = mc²

for intelligence that I've seen.

za inteligenciju koji sam video.

So what you're seeing here

Ono što vidite ovde

is a statement of correspondence

je iskaz odnosa

that intelligence is a force, F,

gde je inteligencija sila, F,

that acts so as to maximize future freedom of action.

koja se ponaša tako da maksimizira buduću slobodu delovanja.

It acts to maximize future freedom of action,

Ponaša se tako da maksimizira buduću slobodu delovanja

or keep options open,

ili drži opcije otvorene,

with some strength T,

sa nekom snagom T,

with the diversity of possible accessible futures, S,

sa raznolikošću mogućih dostupnih budućnosti, S,

up to some future time horizon, tau.

do nekog budućeg vremenskog horizonta, tau.

In short, intelligence doesn't like to get trapped.

Ukratko, inteligencija ne voli da bude zarobljena.

Intelligence tries to maximize future freedom of action

Ona pokušava da maksimizira buduću slobodu delovanja

and keep options open.

i da drži opcije otvorene.

And so, given this one equation,

I tako, kad imate ovu jednačinu,

it's natural to ask, so what can you do with this?

prirodno je pitati šta možete da uradite s njom?

How predictive is it?

Kolika joj je mogućnost predviđanja?

Does it predict human-level intelligence?

Da li predviđa inteligenciju ljudskog nivoa?

Does it predict artificial intelligence?

Da li predviđa veštačku inteligenciju?

So I'm going to show you now a video

Sada ću vam pokazati video

that will, I think, demonstrate

koji će, ja mislim, demonstrirati

some of the amazing applications

neke od neverovatnih primena

of just this single equation.

samo ove jedne jednačine.

(Video) Narrator: Recent research in cosmology

(Video) Narator: Skorija istraživanja u kosmologiji

has suggested that universes that produce

su ukazala da univerzumi koji proizvode

more disorder, or "entropy," over their lifetimes

više nereda ili "entropije" tokom svog postojanja

should tend to have more favorable conditions

bi trebalo da pokazuju tendenciju da imaju povoljnije uslove

for the existence of intelligent beings such as ourselves.

za postojanje inteligentnih bića kao što smo mi.

But what if that tentative cosmological connection

Ali šta ako ta provizorna kosmološka povezanost

between entropy and intelligence

između entropije i inteligencije

hints at a deeper relationship?

nagoveštava dublji odnos?

What if intelligent behavior doesn't just correlate

Šta ako inteligentno ponašanje nije samo u uzajamnoj vezi

with the production of long-term entropy,

sa proizvodnjom dugoročne entropije,

but actually emerges directly from it?

već, u stvari, direktno nastaje od nje?

To find out, we developed a software engine

Da bismo to saznali, razvili smo softver

called Entropica, designed to maximize

zvani Entropika, napravljen da maksimizira

the production of long-term entropy

proizvodnju dugoročne entropije

of any system that it finds itself in.

bilo kog sistema u kome se nađe.

Amazingly, Entropica was able to pass

Entropika je, neverovatno, uspela da prođe

multiple animal intelligence tests, play human games,

višestruke testove životinjske inteligencije, da igra ljudske igre

and even earn money trading stocks,

i čak da zarađuje novac trgujući deonicama,

all without being instructed to do so.

a da joj to nije bilo naloženo.

Here are some examples of Entropica in action.

Evo nekih primera Entropike u akciji.

Just like a human standing upright without falling over,

Baš kao što čovek stoji uspravno, a da ne padne,

here we see Entropica

ovde vidimo Entropiku

automatically balancing a pole using a cart.

koja automatski balansira na gredi koristeći kolica.

This behavior is remarkable in part

Ovo ponašanje je vredno pažnje

because we never gave Entropica a goal.

delom zato što uopšte nismo zadali cilj Entropici.

It simply decided on its own to balance the pole.

Jednostavno je sama odlučila da balansira na gredi.

This balancing ability will have appliactions

Ova sposobnost balansiranja će imati primenu

for humanoid robotics

u humanoidnoj robotici

and human assistive technologies.

i u tehnologiji za pomoć ljudima.

Just as some animals can use objects

Baš kao što neke životinje mogu da koriste predmete

in their environments as tools

u svojoj okolini kao alatke

to reach into narrow spaces,

da bi dohvatili nešto kroz uzak prostor,

here we see that Entropica,

ovde vidimo Entropiku koja,

again on its own initiative,

ponovo na sopstvenu inicijativu,

was able to move a large disk representing an animal

uspeva da pomeri veliki disk koji predstavlja životinju

around so as to cause a small disk,

da bi mali disk

representing a tool, to reach into a confined space

koji predstavlja alatku stigao u zatvoreni prostor

holding a third disk

u kome je treći disk

and release the third disk from its initially fixed position.

i izbacuje treći disk sa njegove početne pozicije.

This tool use ability will have applications

Ova upotreba sposobnosti alatke će imati primene

for smart manufacturing and agriculture.

u "pametnoj" proizvodnji i poljoprivredi.

In addition, just as some other animals

Osim toga, baš kao što su neke druge životinje

are able to cooperate by pulling opposite ends of a rope

sposobne da sarađuju tako što vuku suprotne krajeve konopca

at the same time to release food,

u isto vreme, da bi oslobodile hranu,

here we see that Entropica is able to accomplish

ovde vidimo da je Entropika sposobna da postigne

a model version of that task.

verziju tog zadatka pomoću modela.

This cooperative ability has interesting implications

Ova sposobnost kooperacije ima zanimljive implikacije

for economic planning and a variety of other fields.

na ekonomsko planiranje i mnoštvo drugih oblasti.

Entropica is broadly applicable

Entropika ima široku primenu

to a variety of domains.

na niz domena.

For example, here we see it successfully

Na primer, ovde vidimo kako uspešno igra

playing a game of pong against itself,

igru pong, sama protiv sebe,

illustrating its potential for gaming.

ilustrujući svoj potencijal za igre.

Here we see Entropica orchestrating

Ovde vidimo Entropiku kako pravi

new connections on a social network

nove veze na društvenoj mreži

where friends are constantly falling out of touch

gde prijatelji konstantno gube kontakt

and successfully keeping the network well connected.

i uspešno drži mrežu dobro povezanom.

This same network orchestration ability

Ova sposobnost organizovanja mreže

also has applications in health care,

takođe ima primenu u zdravstvu,

energy, and intelligence.

energetici i inteligenciji.

Here we see Entropica directing the paths

Ovde vidimo Entropiku kako usmerava puteve

of a fleet of ships,

flote brodova,

successfully discovering and utilizing the Panama Canal

tako što uspešno otkriva i koristi Panamski kanal

to globally extend its reach from the Atlantic

da bi ih sprovela od Atlantika

to the Pacific.

do Pacifika.

By the same token, Entropica

Pored toga, Entropika ima

is broadly applicable to problems

široku primenu na probleme

in autonomous defense, logistics and transportation.

autonomne odbrane, logistike i prevoza.

Finally, here we see Entropica

Na kraju, ovde vidimo Entropiku

spontaneously discovering and executing

kako spontano otkriva i izvršava

a buy-low, sell-high strategy

strategiju "kupi jeftino, prodaj skupo"

on a simulated range traded stock,

na simulaciji promena na berzi,

successfully growing assets under management

uspešno uvećavajući dobit pod svojim vođstvom

exponentially.

eksponencijalno.

This risk management ability

Ova sposobnost menadžmenta sa rizikom

will have broad applications in finance

imaće široke primene u finansijama

and insurance.

i osiguranju.

Alex Wissner-Gross: So what you've just seen

Aleks Visner-Gros: Ono što ste upravo videli

is that a variety of signature human intelligent

je niz originalno ljudskih inteligentnih

cognitive behaviors

kognitivnih ponašanja

such as tool use and walking upright

kao što su korišćenje alata i uspravan hod

and social cooperation

i društvena saradnja

all follow from a single equation,

koji proizilaze iz jedne jednačine

which drives a system

koja pokreće sistem

to maximize its future freedom of action.

da bi maksimizirala buduću slobodu delovanja.

Now, there's a profound irony here.

U ovome ima duboke ironije.

Going back to the beginning

Ako se vratimo na početak

of the usage of the term robot,

korišćenja izraza "robot",

the play "RUR,"

dramski komad "R.U.R";

there was always a concept

uvek je postojala ideja

that if we developed machine intelligence,

da ako bismo stvorili mašinsku inteligenciju,

there would be a cybernetic revolt.

desila bi se kibernetska pobuna.

The machines would rise up against us.

Mašine bi se pobunile protiv nas.

One major consequence of this work

Jedna velika posledica ovog rada je ta

is that maybe all of these decades,

da smo možda svih ovih decenija

we've had the whole concept of cybernetic revolt

imali obrnut princip kibernetske

in reverse.

pobune.

It's not that machines first become intelligent

Nije se desilo da su mašine prvo postale inteligentne

and then megalomaniacal

pa onda megalomanijačne

and try to take over the world.

i pokušale da preuzmu svet.

It's quite the opposite,

Potpuno je suprotno.

that the urge to take control

Poriv da se preuzme kontrola

of all possible futures

svih mogućih budućnosti

is a more fundamental principle

je osnovniji princip

than that of intelligence,

od inteligencije.

that general intelligence may in fact emerge

Opšta inteligencija može, u stvari, da nastane

directly from this sort of control-grabbing,

direktno od ovakvog grabljenja kontrole,

rather than vice versa.

pre nego obrnuto.

Another important consequence is goal seeking.

Još jedna važna posledica je traženje cilja.

I'm often asked, how does the ability to seek goals

Često me pitaju kako sposobnost za traženje ciljeva

follow from this sort of framework?

proizilazi iz ovakve strukture?

And the answer is, the ability to seek goals

I odgovor je da će sposobnost za traženje ciljeva

will follow directly from this

proizaći direktno iz ovoga,

in the following sense:

u sledećem smislu:

just like you would travel through a tunnel,

baš kao što biste putovali kroz tunel,

a bottleneck in your future path space,

uskim prolazom u prostoru vašeg budućeg puta,

in order to achieve many other

da biste postigli mnoge druge

diverse objectives later on,

različite ciljeve kasnije

or just like you would invest

ili kao kad biste uložili

in a financial security,

u finansijsku sigurnost,

reducing your short-term liquidity

smanjujući svoju kratkoročnu platežnu moć

in order to increase your wealth over the long term,

da biste uvećali svoje bogatstvo tokom dužeg perioda,

goal seeking emerges directly

traženje ciljeva direktno nastaje

from a long-term drive

od dugoročnog poriva

to increase future freedom of action.

da se uveća buduća sloboda delovanja.

Finally, Richard Feynman, famous physicist,

Na kraju, Ričard Fajnmen, čuveni fizičar,

once wrote that if human civilization were destroyed

jednom je napisao da kad bi ljudska civilizacija bila uništena

and you could pass only a single concept

i kad biste mogli da prenesete samo jednu ideju

on to our descendants

našim potomcima

to help them rebuild civilization,

da im pomognete da ponovo izgrade civilizaciju,

that concept should be

ta ideja bi trebala da bude

that all matter around us

da se sva materija oko nas

is made out of tiny elements

sastoji od sićušnih elemenata

that attract each other when they're far apart

koji se privlače međusobno kad su razdvojeni,

but repel each other when they're close together.

ali se međusobno odbijaju kad su blizu.

My equivalent of that statement

Moj ekvivalent ovoj izjavi

to pass on to descendants

za prenošenje potomcima

to help them build artificial intelligences

da bismo im pomogli da prave veštačke inteligencije

or to help them understand human intelligence,

ili da razumeju ljudsku inteligenciju,

is the following:

je sledeći:

Intelligence should be viewed

inteligenciju bi trebalo posmatrati

as a physical process

kao fizički proces

that tries to maximize future freedom of action

koji pokušava da maksimizira buduću slobodu delovanja

and avoid constraints in its own future.

i izbegne ograničenja u svojoj budućnosti.

Thank you very much.

Hvala vam mnogo.

(Applause)

(Aplauz)